TODO — Next Upstream Interop Lanes (2026 Q2)¶

• Date: 2026-04-08
• Owner: Evidence / Product
• Status: Active queue note. The P11 commerce / trust-proof family is now formally started with PLAN — P11A Visa TAP Intent Verification Evidence Interop, the Browser Use adjacent lane is now live, the PLAN — P13 Langfuse Experiment Result Evidence Interop lane is now sample-backed on main, and P11D APS is explicitly watchlist-only until its promote criteria are met.
• Scope (this document now): Record the ranked post-Agno queue for the next upstream interop lanes, the reasons behind that ordering, and the execution rules learned from the current wave.

2026-05-03 status note: After the v3.6.0-v3.9.1 evidence-receipt line and the P14d Mastra semantic freeze, the next selected candidate is Pydantic, but only through P9c Pydantic Reduced Case-Result Receipt Readiness Freeze. P9b recut the sample around one reduced case-result artifact derived from EvaluationReport.cases[]; P9c freezes the readiness boundary before any importer-only P9d work. ReportCase remains discovery input, not the contract unit.

1. Why this queue exists¶

After the current framework, protocol, runtime-accounting, and eval-report wave, Assay now has enough signal to rank the next outreach candidates more strictly.

The working filter on 2026-04-08 was:

• current repo momentum
• one official seam already documented upstream
• a natural maintainer channel
• a high chance that sample-first outreach will read as a technical boundary question, not as promotion

This note is the queue record for that ranking and the resulting execution order.

It is not an implementation PR.

It is not an outward post.

It is not a commitment to run every candidate below immediately.

2. What changed after the current wave¶

The current wave produced a few useful operating rules:

• GitHub-native outreach works better than forum-first outreach. The LangGraph moderation hold is the clearest counterexample.
• The strongest lanes are the ones where upstream already documents one small official seam.
• Trace-first is no longer the default choice just because a repo exposes observability.
• Platform-adjacent tools require a different posture from framework repos. Import/export slices are usually safer than leading with a generic observability pitch.

Those lessons now drive the queue below.

3. Overall priority list¶

Tier 0 — first finish what is already live¶

Tier 0 means:

• no outward UCP post
• no extra pushes on cold threads
• keep current live lanes breathing unless an upstream maintainer responds

Tier 1 — selected active candidate¶

Tier 2 — parked platform-adjacent candidate¶

Tier 2b — completed importer-only score lane, claim semantics frozen¶

Tier 3 — special-case OTel-native candidate¶

Tier 4 — later frontier and heavier infra lanes¶

P11D promote rule¶

Treat APS as a roadmap watchlist candidate, not an active lane, unless all of the following are true:

1. A smallest external-facing artifact is confirmed publicly in APS materials or in the APS repo itself, for example a signed authorization artifact / receipt that clearly sits below the rest of the protocol stack.
2. The seam can be modeled without pulling in passports, delegation lattice, reputation, or governance primitives.
3. The sample can stay strictly at permit, deny, and malformed.
4. The conversation happens in the APS repo itself, not by continuing the A2A thread or another third-party ecosystem thread.

Tier 5 — still lower fit¶

4. Historical note on Agno¶

At the time of discovery, agno-agi/agno ranked first among the same-space framework candidates because:

• Discussions were enabled
• Evals and Tracing were clearly separated in the docs
• AccuracyEval was a cleaner first seam than another trace-export sample

That choice has now already been executed in the current wave. The formal plan is PLAN — P10 Agno Accuracy Eval Evidence Interop.

That lane is already in motion, so the queue no longer starts with Agno even though Agno remains the strongest general-purpose next-lane choice in the ranking.

2026-05-09 addendum: agno-agi/agno PR #7782 also opened a separate ToolAuditHook JSONL fixture seam. Keep that as a post-merge watchlist / probe-only candidate under the P10 plan. It must not become a new public receipt family, an Agno-specific Harness branch, or a broader tracing / AgentOS lane by default.

5. Historical note on Browser Use¶

browser-use/browser-use was not the highest strategic priority overall, but it was the right adjacent lane to finish before opening the next platform lane because:

• the planning slice was already in progress
• the seam is clean and materially different from the current wave
• it can be finished without opening the heavier P11A commerce branch yet
• it preserves one-lane-at-a-time discipline better than pivoting mid-plan

The critical Browser Use lesson is that the best seam is not observability.

The docs expose:

• AgentHistoryList
• action_history()
• final_result()
• errors()
• structured_output

At the same time, Browser Use also documents Laminar, OpenLIT, and telemetry. That broader observability layer is exactly what Assay should avoid as the first wedge.

That lane is now live. The formal Browser Use plan lives in PLAN — P12 Browser Use History / Output Evidence Interop.

6. Historical note on P11A¶

The P11A Visa TAP lane was ranked above Browser Use in the broader frontier ordering because it had stronger protocol value:

• verification-first rather than platform-first
• cryptographic and protocol-adjacent enough to fit Assay's trust-compiler direction closely
• strategically different from another framework or eval lane

The formal frontier plan now lives in PLAN — P11A Visa TAP Intent Verification Evidence Interop.

That lane is now live too, so the queue no longer needs to choose between Browser Use and P11A as the next move.

7. Why Langfuse is now the next planned lane¶

langfuse/langfuse is now the next best planned lane because the two lanes that previously sat ahead of it in execution order are already in motion.

Why it now moves up:

• strong repo momentum
• Discussions enabled with an answerable Support category
• official eval docs around datasets, experiments, and scores
• API-first and export-friendly positioning
• a seam that is different from Browser Use history/output and TAP verification

Why it is still socially harder than the earlier framework lanes:

• Langfuse already positions itself as a broad LLM engineering platform with observability, datasets, scores, and experiments
• that makes the seam real
• but it also makes the outreach socially riskier because Assay can be read as another platform talking to a platform

The right posture there is:

• export/import sample first
• bounded experiment-result seam first
• Support Discussion only after the sample lands
• no trace-first framing

The formal Langfuse plan now lives in PLAN — P13 Langfuse Experiment Result Evidence Interop.

8. Why Mastra stays below the top four¶

mastra-ai/mastra is slightly lower in strategic weight than Langfuse, but in some respects cleaner to approach because it has less platform-on-platform friction.

The main reason it stays below the top four in this queue is channel shape:

• no Discussions
• outward route is issue-first

That means the lane may eventually be easier socially, but it is less natural as the next GitHub-native sample-backed question.

9. Sequencing rules¶

The queue should be executed under the same discipline as the current wave:

1. one repo at a time
2. sample first
3. one small outward move only after the sample lands on main
4. no second seam in the first sample
5. no observability-first pitch when a smaller result artifact exists

Additional queue rules:

• reserve P11 for the commerce / trust-proof family
• Browser Use should stay output/history-first, not Laminar/OpenLIT-first
• P11A should stay verification-first, not payment-truth-first
• P11D should remain watchlist-only unless a signed authorization artifact can be kept below passports, delegation, reputation, and governance
• Langfuse should stay experiment-result-first, not trace-first
• Mastra should stay eval-result-first, not tracing-first
• OpenLIT should remain a special-case OTel-native candidate, not the default next lane

10. Next actions¶

1. P13 is now plan-backed and sample-backed on main; treat its outward move as the next deliberate platform-adjacent decision, not an automatic rush.
2. Let the fresh Browser Use and Visa TAP outward threads breathe unless an upstream maintainer responds.
3. Keep P11D APS as watchlist-only unless its promote criteria are met in aeoess/agent-passport-system.
4. Keep Mastra as the main fallback if the Langfuse positioning risk feels too high when deciding whether to open the outward Langfuse thread.

References¶

• PLAN — P10 Agno Accuracy Eval Evidence Interop
• PLAN — P11A Visa TAP Intent Verification Evidence Interop
• PLAN — P12 Browser Use History / Output Evidence Interop
• PLAN — P13 Langfuse Experiment Result Evidence Interop